METHOD AND DEVICE FOR SCHEDULING USER REQUEST IN DISTRIBUTED RESOURCE SYSTEM

Abstract

 According to a method for scheduling a user request in a distributed resource system, an apparatus, and a system that are provided by embodiments of the present invention, in a T_n+1 period, an S_d acquires, from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, and the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm. The user z request can be scheduled without depending on a user agent. In addition, the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), the Pⁱ_z,d by using the scheduling algorithm, thereby implementing global scheduling on the user z request and ensuring a performance requirement of the user z.

Description

TECHNICAL FIELD

[0001] The present invention relates to the information technology field, and in particular, to a method for scheduling a user request in a distributed resource system, and an apparatus.

BACKGROUND

[0002] There are multiple resource-providing entities in a distributed resource system. Resources include a computing capability of a central processing unit (Central Processing Unit, CPU), hard disk space, memory space, network bandwidth, and the like, where a resource may be a physical resource or a virtual resource. When a resource-providing entity provides a resource according to a user request, a resource competition occurs between users. Therefore, scheduling a user request to ensure performance for users becomes extremely important. FIG. 1 shows a user request scheduling solution, where a distributed resource system shown in FIG. 1 is a distributed file system and includes a user agent A, a user agent B, a scheduler A, a scheduler B, a resource-providing entity A, and a resource-providing entity B, where the user agent A does not communicate with the user agent B. The user agent A queries a metadata server in the distributed file system according to a user A request, and the user agent A determines, according to information obtained by querying, a scheduler in FIG. 1 to which the user A request is sent. Likewise, the user agent A queries the metadata server in the distributed file system according to a user B request, and the user agent B determines, according to information obtained by querying, a scheduler in FIG. 1 to which the user B request is sent. The resource-providing entity A and the resource-providing entity B are file systems. The resource-providing entity A is configured to provide a resource A, where the resource A is specifically a capability of providing input/output operations per second (Input/Output Operations Per Second, IOPS). The resource-providing entity B is configured to provide a resource B, where the resource B is specifically a capability of providing IOPS. A user A and a user B each send a request to the distributed resource system. For example, the user A sends a user A request to each of the resource-providing entity A and the resource-providing entity B by using the user agent A; then, the resource-providing entity A provides the resource A for the user A request, and the resource-providing entity B provides the resource B for the user A request. The user B sends a user B request to each of the resource-providing entity A and the resource-providing entity B by using the user agent B, and the resource-providing entity A and the resource-providing entity B provide the resources for the user B request. When the user A sends the user A request to each of the resource-providing entity A and the resource-providing entity B, in an implementation manner in FIG. 1, the resources provided by the resource-providing entity A and the resource-providing entity B for the user A are both IOPS capabilities. Likewise, the resources provided by the resource-providing entity A and the resource-providing entity B for the user B are also IOPS capabilities.

[0003] In the distributed resource system shown in FIG. 1, a resource weight is allocated to each of the user A and the user B. For example, a resource weight of the user A is 2, and a resource weight of the user B is 1. A resource weight of a user indicates a resource quota that is allocated by a distributed resource system to the user. The user agent A collects statistics on a quantity of user A requests that are sent by the user A to the scheduler A and the scheduler B, and the user agent B collects statistics on a quantity of user B requests that are sent by the user B to the scheduler A and the scheduler B. The scheduler A allocates, from the resource-providing entity A, a resource for the user A request according to a resource weight of the user A and the quantity of the user A requests that are sent by the user agent A to the scheduler A and the scheduler B. The scheduler A allocates, from the resource-providing entity A, a resource for the user B request according to a resource weight of the user B and the quantity of the user B requests that are sent by the user agent B to the scheduler A and the scheduler B. Likewise, the scheduler B allocates, from the resource-providing entity B, a resource for the user A request according to the resource weight of the user A and the quantity of the user A requests that are sent by the user agent A to the scheduler A and the scheduler B. The scheduler B allocates, from the resource-providing entity B, a resource for the user B request according to the resource weight of the user B and the quantity of the user B requests that are sent by the user agent B to the scheduler A and the scheduler B.

[0004] Because the user agent A and the user agent B are applicable only to a distributed file system scenario, the solution in an architecture shown in FIG. 1 cannot be widely applied to a scenario in which a user request is scheduled in a distributed resource system.

SUMMARY

[0005] According to a first aspect, an embodiment of the present invention provides a method for scheduling a user request in a distributed resource system, where the distributed resource system includes schedulers S_x, resource-providing entities R_x, and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x; the G_y communicates with any S_x; and the method includes:

acquiring, by an S_d in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, where

d and k are natural numbers, ≤1≤d≤M, and 1≤k≤Y; a resource weight of the user z is φ_z; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

scheduling, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm, where the Pⁱ_z,_d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

[0006] With reference to the first aspect of the embodiment of the present invention, in a first possible implementation manner, the scheduling, by the S_d according to φ_z, C_z(T_n), C_z,d (T_n), and N_z,d (T_n), a Pⁱ_z,d by using a scheduling algorithm specifically includes:

computing, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and C_z,d(T_n), a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d ; and adding the Pⁱ_z,d to a scheduling queue, where the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

[0007] With reference to the first possible implementation manner of the first aspect of the embodiment of the present invention, in a second possible implementation manner, S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(Pⁱ_z,d)},

and

where v( Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

[0008] With reference to the first possible implementation manner of the first aspect of the embodiment of the present invention, in a third possible implementation manner, S(Pⁱ_z,d)=max{v(Pⁱ_z,d),



and

where v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

[0009] According to a second aspect, an embodiment of the present invention provides a scheduler S_d, where the scheduler S_d is applied to a distributed resource system, and the distributed resource system includes schedulers S_x, resource-providing entities R_x, and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; d is a number in x; y is consecutive natural numbers from 1 to Y; the S_x communicates with the R_x; a C_y communicates with any S_x ; the scheduler S_d includes a central processing unit and a memory, where the central processing unit executes an executable instruction in the memory, to perform the method in any one of the first aspect to the third possible implementation manner of the first aspect of the embodiment of the present invention.

[0010] According to a third aspect, an embodiment of the present invention provides a scheduler S_d, where the scheduler S_d is applied to a distributed resource system, and the distributed resource system includes schedulers S_x, resource-providing entities R_x and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; d is a number in x; y is consecutive natural numbers from 1 to Y; the S_x communicates with the R_x; a C_y communicates with any S_x ; and the scheduler S_d includes an acquiring unit and a scheduling unit, where:

the acquiring unit is configured to acquire, in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, where

k is a natural number, and 1≤k≤Y; a resource weight of the user z is ø_z; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

the scheduling unit is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), Pⁱ_z,d by using a scheduling algorithm, where Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is resource that is provided by an R_d and consumed by C_z,d(T_n) user z requests received by the S_d in the T_n period.

[0011] According to a fourth aspect, an embodiment of the present invention provides a non-volatile computer readable storage medium, where the non-volatile computer readable storage medium stores a computer instruction that is used to perform user request scheduling in a distributed resource system, and the distributed resource system includes schedulers S_x, resource-providing entities R_x, and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x, and a C_y communicates with any S_x ; and a scheduler S_d executes the computer instruction, to perform the method in any one of the first aspect to the third possible implementation manner of the first aspect of the embodiment of the present invention, where d is a number in x.

[0012] According to a fifth aspect, an embodiment of the present invention provides a distributed resource system, where the distributed resource system includes schedulers S_x, resource-providing entities R_x, and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x; and the G_y communicates with any S_x;

a coordinator G_k of a user z is configured to provide, for a scheduler S_d in a T_n+1 period, a resource C_z(T_n) that is consumed by a user z request in a T_n period, where

d and k are natural numbers, 1≤d≤M, and 1≤k≤Y; a resource weight of the user z is φ_z; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

the scheduler S_d is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a pⁱ_z,d by using a scheduling algorithm, where the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

[0013] According to a sixth aspect, an embodiment of the present invention provides a resource scheduling method in a distributed resource system, where the distributed resource system includes multiple schedulers;

a first scheduler in the multiple schedulers acquires, from a coordinator of the first user, the sum of resources that are consumed in the multiple schedulers by a user request of the first user in a previous period; and

the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

[0014] With reference to the sixth aspect, in a first possible implementation manner, that the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period specifically includes:

computing, by the first scheduler according to the resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, the resource that is consumed in the first scheduler by the user request of the first user in the previous period, and the quantity of the user requests of the first user received by the first scheduler in the previous period, a virtual start time and a virtual finish time of the user request of the first user; and adding the user request of the first user to a scheduling queue, where the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

[0015] According to a seventh aspect, an embodiment of the present invention provides a distributed resource system, where the distributed resource system includes multiple schedulers, and a first scheduler in the multiple schedulers includes an acquiring unit and a scheduling unit, where:

the acquiring unit is configured to acquire, from a coordinator of the first user, the sum of resources that are consumed in the multiple schedulers by a user request of the first user in a previous period; and

the scheduling unit is configured to schedule the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

[0016] With reference to the seventh aspect, in a first possible implementation manner, the scheduling unit is specifically configured to compute, according to the resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, the resource that is consumed in the first scheduler by the user request of the first user in the previous period, and the quantity of the user requests of the first user received by the first scheduler in the previous period, a virtual start time and a virtual finish time of the user request of the first user; and add the user request of the first user to a scheduling queue, where the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

[0017] According to the method for scheduling a user request in a distributed resource system, the apparatus, and the system that are provided by embodiments of the present invention, in a T_n+1 period, an S_d acquires, from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, and the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm. The user z request can be scheduled without depending on a user agent. In addition, the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), the Pⁱ_z,d by using the scheduling algorithm, thereby implementing global scheduling on the user z request and ensuring a performance requirement of the user z.

BRIEF DESCRIPTION OF DRAWINGS

[0018] To describe the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings required for describing the embodiments. The accompanying drawings in the following description show merely some embodiments of the present invention, and other drawings may still be derived from these accompanying drawings.

FIG. 1 is a schematic diagram of a distributed resource system in the prior art;

FIG. 2 is a schematic diagram of a distributed resource system according to an embodiment of the present invention;

FIG. 3a is a diagram of a relationship among a scheduler A, a resource-providing entity A, and a coordinator A;

FIG. 3b is a diagram of a relationship among a scheduler A, a resource-providing entity A, and a coordinator A;

FIG. 4 is a schematic diagram of a method according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a status of a distributed resource system according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of a status of user requests in a scheduling queue;

FIG. 7 is a schematic diagram of a status of user requests in a scheduling queue;

FIG. 8 is a schematic diagram of a status of a distributed resource system according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of a status of user requests in a scheduling queue;

FIG. 10 is a schematic diagram of a status of user requests in a scheduling queue;

FIG. 11 is an architecture diagram of a distributed resource system according to an embodiment of the present invention; and

FIG. 12 is a schematic structural diagram of a scheduler according to an embodiment of the present invention.

DESCRIPTION OF EMBODIMENTS

[0019] The following clearly and completely describes the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are a part rather than all of the embodiments of the present invention. All other embodiments obtained based on the embodiments of the present invention shall fall within the protection scope of the present invention.

[0020] As shown in FIG. 2, an embodiment of the present invention provides a distributed resource system. The distributed resource system includes a scheduler A, a scheduler B, a resource-providing entity A, a resource-providing entity B, a coordinator A, and a coordinator B. The scheduler A communicates with the resource-providing entity A, and is configured to schedule a user request sent to the resource-providing entity A. The scheduler B communicates with the resource-providing entity B, and is configured to schedule a user request sent to the resource-providing entity B. The coordinator A communicates with the scheduler A and the scheduler B, and the coordinator B communicates with the scheduler A and the scheduler B. A resource in this embodiment of the present invention may be a physical resource or a virtual resource, such as a CPU computing capability, hard disk space, a memory space, network bandwidth, or the like. A user in this embodiment of the present invention may refer to a client, a virtual machine, an application process, or the like, which is not specifically limited in this embodiment of the present invention. A resource allocated by a resource-providing entity for a user A request and a resource allocated by the resource-providing entity for a user B request are resources of a same type. Resources that are provided by the resource-providing entity A and the resource-providing entity B for a user A request and resources that are provided by the resource-providing entity A and the resource-providing entity B for a user B request are resources of a same type. In this embodiment of the present invention, a resource allocated by a resource-providing entity to a user request has the same meaning as a resource provided by a resource-providing entity for a user request, which may also be expressed as a resource of a resource-providing entity that is consumed by a user request, or a resource that is provided by a resource-providing entity and consumed by a user request. A resource that is provided by a resource-providing entity and consumed by a user request refers to a resource that needs to be consumed by the user request, which specifically includes a resource that has been consumed, or includes, if the user request has already been processed by the resource-providing entity, a resource that needs to be consumed by the user request after the user request is received by a scheduler and before the user request is processed by the resource-providing entity. In an implementation manner, after receiving a user request, a scheduler can calculate a quantity of resources that are provided by a resource-providing entity and need to be consumed by the user request. For example, if a scheduler A receives 100 user requests in a period, resources that are provided by a resource-providing entity and consumed by the 100 user requests can be calculated by summing up resources that are provided by the resource-providing entities and need to be consumed by all user requests.

[0021] This embodiment of the present invention is described by using an example in which a resource is an IOPS capability, the resource-providing entity A is a storage array A, and the resource-providing entity B is a storage array B.

[0022] Before this embodiment of the present invention is further described, it should be noted that, a relationship among the scheduler A, the coordinator A, and the resource-providing entity A shown in FIG. 2 is just a logical representation. In specific implementation, the resource-providing entity A, the scheduler A, and the coordinator A may be located in a same apparatus, as shown in FIG. 3a; or, the resource-providing entity A and the scheduler A are located in a same apparatus while the coordinator A is located independently of the resource-providing entity A and the scheduler A, as shown in FIG. 3b; or, another implementation manner may be used. A specific implementation manner is not limited herein. Likewise, for a relationship among the scheduler B, the coordinator B, and the resource-providing entity B shown in FIG. 2, refer to FIG. 3a and FIG. 3b. The scheduler may be an independent device or a hardware module. A function of the scheduler may be further implemented by a processor by executing a specific computer instruction. For example, if functions of the scheduler A, the coordinator A, and the resource-providing entity A are all implemented by a processor by executing specific computer instructions, communication between the scheduler A and the coordinator A, and communication between the scheduler A and the resource-providing entity A may be presented as instruction invoking or links between different program modules. Likewise, the coordinator may be an independent device or a hardware module. A function of the coordinator may be further implemented by a processor by executing a specific computer instruction, which is not limited in this embodiment of the present invention. In specific implementation, steps of the method implemented by a scheduler in this embodiment of the present invention may be implemented by a computer processor by executing a computer instruction in a memory. Details are not described herein again.

[0023] In the distributed resource system shown in FIG. 2, when a user A sends a user A request to the storage array A and the storage array B, both the storage array A and the storage array B need to consume an IOPS capability when processing the user A request. Likewise, when a user B sends a user B request to the storage array A and the storage array B, IOPS capabilities of the storage array A and the storage array B also need to be consumed. Particularly, when the IOPS capability of the storage array A cannot meet a requirement of processing the user A request and the user B request at the same time, the scheduler A needs to schedule the received user A request and the received user B request according to a scheduling algorithm; when the IOPS capability of the storage array B cannot meet a requirement of processing the user A request and the user B request at the same time, the scheduler B needs to schedule the received user A request and the received user B request according to a scheduling algorithm. There is a resource competition relationship between the user A and the user B. To ensure performance for the users, a resource weight is generally allocated to each user in the distributed resource system. The user A and the user B are used as an example in this embodiment of the present invention, where a resource weight φ_A, of the user A is 2, and a resource weight φ_B of the user B is 1. Then a ratio of the resource weight of the user A to the resource weight of the user B is 2:1. The resource weights of the user A and the user B are configured in both the scheduler A and the scheduler B. Specifically, the resource weight of the user A may be delivered by the user A to each of the scheduler A and the scheduler B, and the resource weight of the user B is delivered by the user B to each of the scheduler A and the scheduler B; the scheduler A and the scheduler B perform the foregoing configuration. A form of a resource weight is not limited to the form described in this embodiment of the present invention, and may also be that the resource weight φ_A of the user A is 2/3, and the resource weight φ_B of the user B is 1/3.

[0024] The scheduler A receives the user A request sent by the user A, and the scheduler A receives the user B request sent by the user B. The storage array A needs to provide the IOPS capability for both the user A request and the user B request, or in other words, both the user A request and the user B request consume the IOPS capability provided by the storage array A. Likewise, the scheduler B receives the user A request sent by the user A, and the scheduler B receives the user B request sent by the user B. When the IOPS capability of the storage array A can meet the user A request and the user B request, the scheduler A does not need to schedule the received user A request and the received user B request according to a scheduling algorithm; when the IOPS capability of the storage array B can meet the user A request and the user B request, the scheduler B does not need to schedule the received user A request and the received user B request according to a scheduling algorithm. The scheduler A only needs to compute virtual start times and virtual finish times of the received user requests according to a scheduling algorithm. Likewise, the scheduler B also only needs to compute virtual start times and virtual finish times of the received user requests according to a scheduling algorithm. For meanings of a virtual start time and a virtual finish time, refer to a start-time fairness queuing (Start-time fairness queuing, SFQ) algorithm. When the user requests need to be scheduled, the scheduler A needs to add, according to a scheduling algorithm, the received user A request and the received user B request to a scheduling queue of the scheduler A for scheduling; the scheduler B also needs to add, according to the scheduling algorithm, the received user A request and the received user B request to a scheduling queue of the scheduler B for scheduling.

[0025] As shown in FIG. 4, in step 401, a scheduler A acquires, in a T_n+1 period and from a home coordinator A of a user A, a resource C_A(T_n) that is consumed by a user A request in a T_n period, and acquires, from a home coordinator B of a user B, a resource C_B(T_n) that is consumed by a user B request in the T_n period, where T_n indicates the n^th period, and T_n+1 indicates the (n+1)^th period; C_A(T_n) indicates the sum of a resource C_A,A(T_n) that is provided by a resource-providing entity A and consumed by C_A,A(T_n) user A requests received by the scheduler A in the T_n period, and a resource C_A,B(T_n) that is provided by a resource-providing entity B and consumed by N_A,B(T_n) user A requests received by a scheduler B in the T_n period; and C_B(T_n) indicates the sum of a resource C_B,A(T_n) that is provided by the resource-providing entity A and consumed by N_B,A(T_n) user B requests received by the scheduler A in the T_n period, and a resource C_B,B(T_n) that is provided by the resource-providing entity B and consumed by N_B,B(T_n) user B requests received by the scheduler B in the T_n period. The home coordinator A of the user A refers to a coordinator that acquires, in a distributed resource system, the sum of a resource of the storage array A that is consumed by the user A request processed by the scheduler A in the T_n period, and a resource of the storage array A that is consumed by the user A request processed by the scheduler B in the T_n period. The home coordinator A of the user A is also called a coordinator A of the user A. Likewise, a home coordinator of a user is also called a coordinator of the user.

[0026] The scheduler A acquires, in the T_n period and according to a resource that is provided by the resource-providing entity A and consumed by each received user A request, the resource C_A,A(T_n) that is provided by the resource-providing entity A and consumed by the N_A,A(T_n) user A requests processed by the scheduler A in the T_n period; and the scheduler A acquires, in the T_n period and according to a resource that is provided by the resource-providing entity A and consumed by each received user B request, the resource C_B,A(T_n) that is provided by the resource-providing entity A and consumed by the N_B,A(T_n) user B requests processed by the scheduler A in the T_n period. Likewise, the scheduler B acquires, in the T_n period and according to a resource that is provided by the resource-providing entity B and consumed by each received user A request, the resource C_A,B(T_n) that is provided by the resource-providing entity B and consumed by the N_A,B(T_n) user A requests processed by the scheduler B in the T_n period; and the scheduler B acquires, in the T_n period and according to a resource that is provided by the resource-providing entity B and consumed by each received user B request, the resource C_B,B(T_n) that is provided by the resource-providing entity B and consumed by the N_B,B(T_n) user B requests processed by the scheduler B in the T_n period. Resources consumed by the user A requests processed in the T_n period are C_A(T_n)=C_A,A(T_n)+C_A,B(T_n), and resources consumed by the user B requests processed in the T_n period are C_B(T_n)=C_B,A(T_n)+C_B,B(T_n).

[0027] Specifically, the coordinator A acquires C_A,A(T_n) and C_A,B(T_n). An implementation manner is that the scheduler A proactively sends the C_A,A(T_n) to the coordinator A, and the scheduler B proactively sends the C_A,B(T_n) to the coordinator A. Another implementation manner is that the coordinator A requests to acquire the C_A,A(T_n) from the scheduler A, and the coordinator A requests to acquire the C_A,B(T_n) from the scheduler B. Specifically, the coordinator A may acquire the C_A,A(T_n) and the C_A,B(T_n) at one time, or the coordinator A may communicate with the scheduler A and the scheduler B in real time, to acquire, in real time, the resource that is provided by the resource-providing entity A and consumed by each user A request received by the scheduler A, and acquire, in real time, the resource that is provided by the resource-providing entity B and consumed by each user A request received by the scheduler B. For a manner in which the coordinator B acquires the resources C_B(T_n) consumed for processing the user B requests, refer to the manner of the coordinator A, and details are not described herein again.

[0028] The scheduler A acquires, from the home coordinator A of the user A, the resource C_A(T_n) consumed by the user A requests in the T_n period, and acquires, from the home coordinator B of the user B, the resource C_B(T_n) consumed by the user B requests in the T_n period. In an implementation manner, in a T_n+1 period, the coordinator A sends the C_A(T_n) to each of the scheduler A and the scheduler B, and the coordinator B sends the C_B(T_n) to each of the scheduler A and the scheduler B. In another implementation manner, in a T_n+1 period, the scheduler A requests the C_A(T_n) from the coordinator A, the scheduler A requests the C_B(T_n) from the coordinator B; and the scheduler B requests the C_A(T_n) from the coordinator A, the scheduler B requests the C_B(T_n) from the coordinator B. The foregoing two manners are both described as: in a T_n+1 period, the scheduler A acquires, from the home coordinator A of the user A, the resources C_A(T_n) consumed by the user A requests in the T_n period, and acquires, from the home coordinator B of the user B, the resources C_B(T_n) consumed by the user B requests in the T_n period. It can be known from the foregoing description that, in either of the foregoing implementation manners, the scheduler A stores the N_A,A(T_n), N_B,A(T_n), C_A,A(T_n), and C_B,A(T_n); and the scheduler B stores the N_A,B(T_n), N_B,B(T_n), C_A,B(T_n), and C_B,B(T_n).

[0029] As for determining of a home coordinator of a user, a manner may be that a coordinator is configured as a home coordinator of a user; or when an identifier of a user is an integer, a modulo operation is performed on a total quantity of coordinators according to the identifier of the user, to determine a home coordinator of the user; or a hash function may be used to perform computation on an identifier of a user to obtain a hash value, and then a modulo operation is performed on a total quantity of coordinators by using the hash value, to determine a home coordinator of the user. For example, there are 100 users and 20 coordinators, and identifiers of the users are from 1 to 100, the 20 coordinators are numbered from 1 to 20, and a modulo operation is performed on a total quantity of the coordinators according to the identifiers of the users, to determine a home coordinator of each user. The present invention sets no limitation thereto. A total quantity of coordinators may be less than or equal to a quantity of schedulers, and each coordinator can communicate with any scheduler.

[0030] Step 402. The scheduler A schedules, according to φ A, C_A(T_n), C_A,A(T_n), and N_A,A(T_n), the received user A request by using a scheduling algorithm; and the scheduler A schedules, according to φ _B, C_B(T_n), C_B,A(T_n), and N_B,A(Tn), the received user B request by using the scheduling algorithm.

[0031] Likewise, the scheduler B schedules, according to φ A, C_A(T_n), C_A,B(T_n), and N_A,B(T_n), a received user A request by using a scheduling algorithm; and the scheduler B schedules, according to φ _B, C_B(T_n), C_B,B(T_n), and N_B,B(T_n), a received user B request by using the scheduling algorithm.

[0032] When a user request in a scheduler is scheduled according to the method shown in FIG. 4, the scheduler A performs scheduling according to a quantity of resources that are consumed by the user A request in the distributed system in a previous period, a resource weight of the user A, a quantity of user A requests received by the scheduler A in the previous period, and the resource that is provided by the resource-providing entity A and consumed by the user A request received by the scheduler A in the previous period, where the quantity of resources that are consumed by the user A request in the distributed system in a previous period is acquired from the coordinator A. In an architecture without a user agent in the prior art, a coordinator is used to acquire a quantity of resources that are consumed by a user A request in a distributed system in a previous period, so that a scheduler A does not need to perform an extra operation, which can reduce a resource usage of the scheduler A. In addition, the scheduler schedules the user A request according to φ A, C_A(T_n), C_A,A(T_n), and N_A,A(T_n), thereby implementing global scheduling on the user A requests and ensuring a performance requirement of a user z.

[0033] When the scheduler A schedules the user B request, or the scheduler B schedules the user A request, or the scheduler B schedules the user B request, the same foregoing effect can also be achieved. In a user agent-based architecture in the prior art, a scheduler can acquire only a user request and statistics of the user request that are sent by one user agent; therefore, it is difficult to implement global scheduling on a same user request.

[0034] In this embodiment of the present invention, that a scheduler A schedules a user A request is used as an example. During global scheduling on the user A request, the scheduler A needs to consider C_A(T_n) in addition to φ _A, C_A,A(T_n), and N_A,A(T_n) in a previous period, thereby implementing scheduling of the user A request and further ensuring performance for a user A.

[0035] Specifically, in this embodiment of the present invention, Pⁱ_A,A indicates the i^th user A request that is sent by the user A and received by the scheduler A, where a storage array A provides a resource for the Pⁱ_A,A; and Pⁱ_A,B indicates the i^th user A request that is sent by the user A and received by a scheduler B, where a storage array B provides a resource for the Pⁱ_A,B. P^k_B,A indicates the k^th user B request that is sent by a user B and received by the scheduler A, where the storage array A provides a resource for the P^k_B,A ; and P^k_B,B indicates the k^th user B request that is sent by the user B and received by the scheduler B, where the storage array B provides a resource for the P^k_B,B. In this embodiment of the present invention, φ_A=2, and φ_B=1.

[0036] In a T_n period, when a quantity of user A requests and user B requests that are received by the scheduler A is less than an IOPS capability of the storage array A, there is no need to schedule, by using a scheduling algorithm, the user A requests and the user B requests that are received by the scheduler A, and the user A requests and the user B requests are directly processed by the storage array A. When a quantity of user A requests and user B requests that are received by the scheduler B is less than an IOPS capability of the storage array B, there is also no need to schedule, by using a scheduling algorithm, the user A requests and the user B requests that are received by the scheduler B, and the user A requests and the user B requests are directly processed by the storage array B. However, a home coordinator A of the user A needs to acquire resources that are consumed by user A requests received by schedulers in the T_n period. In this embodiment of the present invention, the home coordinator A of the user A acquires a resource C_A(T_n) that is consumed by the user A requests that are received by the scheduler A and the scheduler B in the T_n period. A home coordinator B of the user B also acquires a resource C_B(T_n) that is consumed by the user B requests that are received by the scheduler A and the scheduler B in the T_n period. The coordinator A sends C_A(T_n) to the scheduler A and the scheduler B, and the coordinator B sends C_B(T_n) to the scheduler A and the scheduler B. When a quantity of the user A requests received by the scheduler A in the T_n period is N_A,A(T_n), a consumed resource that is provided by the storage array A is C_A,A(T_n); when a quantity of the user B requests received by the scheduler A in the T_n period is N_B,A(T_n), a consumed resource that is provided by the storage array A is C_B,A(T_n). When a quantity of the user A requests received by the scheduler B in the T_n period is N_A,B(T_n), a consumed resource that is provided by the storage array B is C_A,B(T_n); when a quantity of the user B requests received by the scheduler B in the T_n period is N_B,B(T_n), a consumed resource that is provided by the storage array A is C_B,B(T_n).

[0037] The scheduler A is used as an example. In a T_n+1 period, the scheduler A acquires, from the home coordinator A of the user A, a resource C_A(T_n) that is consumed by a user A request in the T_n period. In the T_n+1 period, when an IOPS capability of the storage array A cannot meet a user A request and a user B request, the scheduler A adds, by using a scheduling algorithm, the received user A request and the received user B request to a scheduling queue of the scheduler A for scheduling. When receiving the user A request and the user B request, the scheduler A computes a virtual start time and a virtual finish time of the user A request, and a virtual start time and a virtual finish time of the user B request. The user A request and the user B request are added to the scheduling queue, and the user A request and the user B request are ranked in the scheduling queue in ascending order of the virtual start times for processing. A user request whose virtual start time is the earliest is ranked at the queue head of the scheduling queue, and a user request whose virtual start time is the latest is ranked at the queue tail of the scheduling queue. The queue head of the scheduling queue refers to a position at which the storage array A performs user request processing first in the scheduling queue. When processing the user requests, the storage array A acquires a user request from the queue head of the scheduling queue for processing, that is, a user request whose virtual start time is the earliest is processed first. The storage array A processes the user requests in the scheduling queue in ascending order of the virtual start times of the user requests.

[0038] In a first implementation manner, a virtual start time of a Pⁱ_A,A that is received by the scheduler A is represented by S(Pⁱ_A,A), and a virtual finish time of the Pⁱ_A,A is represented by F(Pⁱ_A,A). S(Pⁱ_A,A)=max{v(Pⁱ_A,A), F(P^i-1_A,A)}, where v(Pⁱ_A,A)indicates a virtual time of the scheduler A when the scheduler A receives the Pⁱ_A,A, F(P^i-1_A,A)indicates a virtual finish time of a P^i-1_A,A, and max{v(Pⁱ_A,A), F(P^i-1_A,A)} indicates a maximum value in v(Pⁱ_A,A) and

For meanings of the virtual start time S(Pⁱ_A,A), the virtual finish time F(Pⁱ_A,A), and the virtual time v(Pⁱ_A,A), refer to a start-time fairness queuing (Start-time fairness queuing, SFQ) algorithm, and details are not described in this embodiment of the present invention again. c(Pⁱ_A,A) indicates a resource that is provided by the storage array A and consumed by the Pⁱ_A,A, which is one IOPS in this embodiment of the present invention; and d(P_A,A(T_n+1)) indicates a value of a delay of each user A request that is received by the scheduler A in a T_n+1 period, and

Likewise, a virtual start time of a P^k_B,A that is received by the scheduler A is represented by S(P^k_B,A), and a virtual finish time of the P^k_B,A is represented by F(P^k_B,A). S(P^k_B,A)=max {v(P^k_B,A), F(P^k-1_B,A)}, where v(P^k_B,A) indicates a virtual time of the scheduler A when the scheduler A receives the P^k_B,A, F(p^k-1_B,A) indicates a virtual finish time of a P^k-1_B,A, and max {v(P^k_B,A), F(P^k-1_B,A)} indicates a maximum value in v(P^k_B,A) and

where c(P^k_B,A) indicates a resource that is provided by the storage array B and consumed by the P^k _B,A, which is one IOPS in this embodiment of the present invention; and d(P_B,A(T_n+1)) indicates a value of a delay of each user B request that is received by the scheduler A in a T_n+1 period, and

[0039] A virtual start time of a Pⁱ_A,B that is received by the scheduler B is represented by S(Pⁱ_A,B), and a virtual finish time of the Pⁱ_A,B is represented by F(Pⁱ_A,B).S(Pⁱ_A,B)=max{v(Pⁱ_A,B), F(P^i-1_A,B)}, where v(Pⁱ_A,B) indicates a virtual time of the scheduler B when the scheduler B receives the Pⁱ_A,B, F(P^i-1_A,B) indicates a virtual finish time of a P^i-1_A,B, and max {v(Pⁱ_A,B), F(Pⁱ_A,B)} indicates a maximum value in v(Pⁱ_A,B) and F(P^i-1_A,B).

where c(Pⁱ_A,B) indicates a resource that is provided by the storage array B and consumed by the Pⁱ_A,B, which is one IOPS in this embodiment of the present invention; and d(P_A,B(T_n+1)) indicates a value of a delay of each user A request that is received by the scheduler B in a T_n+1 period, and

Likewise, a virtual start time of a P^k_B,B that is received by the scheduler B is represented by S(P^k_B,B), and a virtual finish time of the P^k_B,B is represented by F(P^k_B,B). S(P^k_B,B)=max{v(P^k_B,B), F(P^k-1_B,B)}, where v(P^k_B,B) indicates a virtual time of the scheduler B when the scheduler B receives the Pⁱ_B,B, F(P^k-1_B,B) indicates a virtual finish time of a P^k-1_B,B, and max {v(P^k_B,B), F(P^k-1_B,B)} indicates a maximum value in v(P^k _B,B) and F(P^k-1 _B,B).

where c(P^k _B,B) indicates a resource that is provided by the storage array B and consumed by the P^k_B,B, which is one IOPS in this embodiment of the present invention; and d(P_B,B (T_n+1)) indicates a value of a delay of each user B request that is received by the scheduler B in a T_n+1 period, and

[0040] In a second implementation manner, a virtual start time of a Pⁱ_A,A that is received by the scheduler A is represented by S(Pⁱ_A,A), and a virtual finish time of the Pⁱ_A,A is represented by F(pⁱ_A,A). S(pⁱ_A,A)=max {v(pⁱ _A,A),

}, where v(Pⁱ_A,A)indicates a virtual time of the scheduler A when the scheduler A receives the Pⁱ_A,A; d(P_A,A(T_n+1)) indicates a value of a delay of each user A request that is received by the scheduler A in a T_n+1 period, and

F(P^i-1_A,A) indicates a virtual finish time of a P^i-1_A,A; and max {v(Pⁱ_A,A),

} indicates a maximum value in v(Pⁱ_A,A)and

For meanings of the virtual start time S(Pⁱ_A,A), the virtual finish time F(Pⁱ_A,A) and the virtual time v(Pⁱ_A,A), refer to a start-time fairness queuing (Start-time fairness queuing, SFQ) algorithm, and details are not described in this embodiment of the present invention again. c(Pⁱ_A,A) indicates a resource that is provided by the storage array A and consumed by the Pⁱ_A,A, which is one IOPS in this embodiment of the present invention. Likewise, a virtual start time of a P^k_B,A that is received by the scheduler A is represented by S(P^k_B,A), and a virtual finish time of the P^k_B,A is represented by F(P^k_B,A). S(P^k_B,A)=max {v(P^k_B,A),

}, where v(P^k_B,A) indicates a virtual time of the scheduler A when the scheduler A receives the P^k_B,A; d(P_B,A(T)) indicates a value of a delay of each user B request that is received by the scheduler A in a T_n+1 period, and

F(P^k-1_B,A) indicates a virtual finish time of a P^k-1_B,A; and max{v(P^k_B,A),

} indicates a maximum value in v(P^k_B,A ) and



where c(P^k_B,A) indicates a resource that is provided by the storage array B and consumed by the P^k_B,A, which is one IOPS in this embodiment of the present invention.

[0041] A virtual start time of a Pⁱ_A,B that is received by the scheduler B is represented by S(Pⁱ_B,A), and a virtual finish time of the Pⁱ_A,B is represented by F(Pⁱ_A,B). S(Pⁱ_A,B)=max{v(Pⁱ_A,B),

}, where v(Pⁱ_A,B) indicates a virtual time of the scheduler B when the scheduler B receives the Pⁱ_A,B; d(P_A,B(T_n+1)) indicates a value of a delay of each user A request that is received by the scheduler B in a T_n+1 period, and

F(P^i-1_A,B) indicates a virtual finish time of a P^i-1_A,B; and max{v(Pⁱ_A,B),

indicates a maximum value in v(Pⁱ_A,B) and

where c(Pⁱ_A,B) indicates a resource that is provided by the storage array B and consumed by the Pⁱ_A,B, which is one IOPS in this embodiment of the present invention. Likewise, a virtual start time of a P^k_B,B that is received by the scheduler B is represented by S(P^k_B,B), and a virtual finish time of the P^k_B,Bis represented by F(P^k_B,B). S(P^k_B,B)=max{v(P^k_B,B),

}, where v(P^k_B,B) indicates a virtual time of the scheduler B when the scheduler B receives the P^k_B,B; d(P_B,B(T_n+1)) indicates a value of a delay of each user B request received by the scheduler A in a T_n+1 period, and

F(P^k-1_B,B) indicates a virtual finish time of a P^k-1_B,B; and max{v(P^k_B,B),

} indicates a maximum value in v(P^k_B,B) and

where c(P^k_B,B) indicates a resource that is provided by the storage array B and consumed by the P^k_B,B, which is one IOPS in this embodiment of the present invention.

[0042] The first manner of computing a virtual start time and a virtual finish time of a user request is used as an example. As shown in FIG. 5, in this embodiment, after a distributed resource system is initialized, user A requests Pⁱ_A,A that are received by a scheduler A in a T_n+1 period are Pⁱ_A,A,P²_A,A, P³_A,A, P⁴_A,A, and P⁵_A,A respectively, 5 in total. A quantity of user A requests Pⁱ_A,B that are received by a scheduler B is 0. User B requests P^k_B,A that are received by the scheduler A are P¹_B,A, P²_B,A, P³_B,A, and P⁴_B,A respectively, 4 in total.

[0043] User B requests P^k_B,B that are received by the scheduler B are P¹_B,B, P²_B,B, P³_B,B, P⁴_B,B, and P⁵_B,B respectively, 5 in total. In this embodiment of the present invention, the T_n+1 period is actually the first period. The Pⁱ_A,A is the first user A request received by the scheduler A. In this case, the received Pⁱ_A,A needs to be added to a scheduling queue of the scheduler A for scheduling according to a scheduling algorithm. First, a virtual start time S(P¹_A,A)=max{v(P¹_A,A), F(P^i-1_A,A)} of the Pⁱ_A,A is computed. Because the T_n+1 period is actually the first period, that is, the scheduler A does not receive any Pⁱ_A,A previously, F(P⁰_A,A) is Null, the P¹_A,A is received after initialization of the scheduler A is completed, and S(Pⁱ_A,A )=v(Pⁱ_A,A)=0. A virtual finish time of the P¹_A,A is computed:

where S(P¹_A,A)=0, c(Pⁱ_A,A)=1, φ_A =2,

c_A,B(T_n) =0, and N_A,A(T_n) =0; then d(P_A,A(T_n+1))=0, and

Likewise, a virtual start time of the P²_A,A is 0.5, and a virtual finish time of the P²_A,A is 1; a virtual start time of the P³_A,A is 1, and a virtual finish time of the P³_A,A is 1.5; a virtual start time of the P⁴_A,A is 1.5, and a virtual finish time of the P⁴_A,A is 2; and a virtual start time of the P⁵_A,A is 2, and a virtual finish time of S_x ; the P⁵_A,A is 2.5

in FIG. 5 indicates that the virtual start time of the P¹_A,A is 0 and the virtual finish time of the P¹_A,A is 0.5. For other similar identifiers in FIG. 5 and subsequent figures, refer to the meaning of

and details are not described herein again.

[0044] For the user B request P¹_B,A that is received by the scheduler A, a virtual start time of the P¹_B,A is computed: S(P¹_B,A)=max{v(P¹_B,A), F(P⁰_B,A)}. Because the T_n+1 period is actually the first period, that is, the scheduler A does not receive any P^k_B,A previously, F(P⁰_B,A) is empty, the P¹_B,A is received after initialization of the scheduler A is completed, and v(P¹_B,A)=0. A virtual finish time of the P¹_B,A is

where c(P¹_B,A)=1,

C_B,B(T_n) =0, and N_B,A(T_n)=0; then d(P_B,A(T_n+1))=0, φ_B=1, and F(P¹_B,A)=1 Likewise, a virtual start time of the P²_B,A is 1, and a virtual finish time of the P²_B,A is 2; a virtual start time of the P³_B,A is 2, and a virtual finish time of the P³_B,A is 3; and a virtual start time of the P⁴_B,A is 3, and a virtual finish time of the P⁴_B,A is 4.

[0045] In this case, for the user B request P⁴_B,B that is received by the scheduler B, the received P¹_B,B needs to be scheduled, and a virtual start time of the P¹_B,B is computed: S(P¹_B,B)=max{v(P¹_B,B), F(P⁰_B,B)}. Because the T_n+1 period is actually the first period, that is, the scheduler B does not receive any P^k_B,B previously, F(P⁰_B,B) is empty, the P¹_B,B is received after initialization of the scheduler B is completed, and v(P¹_B,B)=0. A virtual finish time of the B_B1 is

where c(P¹_B,B)=1,

C_B,A(T_n) =0, and N_B,B(T_n) =0; then d(P_B,B(T_n+1))=0, φ_B =1, and F(P¹_B,B)=1. Likewise, a virtual start time of the P²_B,B is 1, and a virtual finish time of the P²_B,B is 2; a virtual start time of the P³_B,B is 2, and a virtual finish time of the P³_B,B is 3; a virtual start time of the P⁴_B,B is 3, and a virtual finish time of the P⁴_B,B is 4; and a virtual start time of the P⁵_B,B is 4, and a virtual finish time of the P⁵_B,B is 5.

[0046] The scheduler A computes virtual start times and virtual finish times of the P¹_A,A, P²_A,A P³_A,A, P⁴_A,A, and P⁵_A,A, and computes virtual start times and virtual finish times of the P¹_B,A, P²_B,A, P³_B,A, and P⁴_B,A. According to the virtual start times of the user requests that are received by the scheduler A, the user requests are ranked in a scheduling queue in ascending order of the virtual start times for processing, that is, an order from the queue head to the queue tail in the scheduling queue is P¹_A,A, P²_A,A, P¹_B,A, P³_A,A, P⁴_A,A, P²_B,A, P⁵_A,A, P³_B,A, and P⁴_B,A, as shown in FIG. 6. The storage array A processes the user requests in the order of P¹_A,A, P²_A,A, P¹_B,A, P³_A,A, P⁴_A,A, P²_B,A, P⁵_A,A, P³_B,A, and P⁴_B,A As shown in FIG. 6, when there is resource competition between the user A requests and the user B requests, a group of user requests P¹_A,A, P²_A,A, and P¹_B,A in the scheduling queue consume, according to a ratio of 2:1, the IOPS that is provided by the storage array A. Another group of user requests P³_A,A, P⁴_A,A, and P²_B,A in the scheduling queue also meets the requirement.

[0047] The scheduler B computes virtual start times and virtual finish times of the P¹_B,B, P²_B,B, P³_B,B, P⁴_B,B, and P⁵_B,B respectively. According to the virtual start times of all the user requests that are received by the scheduler B, the user requests are ranked in a scheduling queue in ascending order of the virtual start times for processing, that is, an order from the queue head to the queue tail in the scheduling queue is P¹_B,B, P²_B,B, P³_B,B, P⁴_B,B, and P⁵_B,B B_B5, as shown in FIG. 7. The storage array B processes the user requests in the order of P¹_B,B, P²_B,B, P³_B,B, P⁴_B,B, and P⁵_B,B. Because there are only the user B requests in the scheduler B, there is no resource competition, and the user B requests are processed in ascending order of the virtual start times of the user B requests.

[0048] A person skilled in the art should understand that, in specific implementation, a quantity of user requests that are received by the scheduler A and the scheduler B in a T_n+1 period may be much greater than the quantity of the user requests provided in the embodiment. A quantity of users is not limited to two. For ease of description in this embodiment of the present invention, in the T_n+1 period, the user A requests that are received by the scheduler A are P¹_A,A, P²_A,A, P³_A,A, P⁴_A,A, and P⁵_A,A, and the user B requests that are received by the scheduler A are P¹_B,A, P²_B,A, P³_B,A, and P⁴_B,A; and the user B requests that are received by the scheduler B are P¹_B,B, P²_B,B, P³_B,B, P⁴_B,B, and P⁵_B,B.

[0049] According to the manner described above, the scheduler A computes a resource C_A,A(T_n+1) that is provided by the storage array A and consumed by N_A,A(T_n+1) user A requests received in the T_n+1 period, and a resource C_B,A(T_n) that is provided by the storage array A and consumed by N_B,A(T_n+1) received user B requests. The scheduler A computes a resource C_A,A(T_n+1) that is provided by the storage array A and consumed by N_A,A(T_n+1) user A requests received in the T_n+1 period, which is specifically that: in this embodiment of the present invention, one user A request consumes one IOPS, and a quantity of the user A requests received by the scheduler A in the T_n+1 period is N_A,A(T_n+1) =5, and the consumed resource C_A,A(T_n+1) that is provided by the storage array A is five IOPSs. Likewise, a quantity of the user B requests that are received by the scheduler A in the T_n+1 period is N_B,A(T_n+1)=4, and the consumed resource C_B,A(T_n+1) that is provided by the storage array A is four IOPSs. A quantity of the user A requests that are received by the scheduler B in the T_n+1 period is N_A,B(T_n+1)=0, and the consumed resource C_A,B(T_n+1) that is provided by the storage array B is 0 IOPS. Likewise, a quantity of the user B requests that are received by the scheduler B in the T_n+1 period is N_B,B(T_n+1)=5, and the consumed resource C_B,B(T_n+1) that is provided by the storage array B is five IOPSs. The coordinator A acquires that the resource C_A,A(T_n+1) that is provided by the storage array A and consumed by the user A requests processed by the scheduler A in the T_n+1 period is five IOPSs, and the resource that is provided by the storage array B and consumed by the user A requests processed by the scheduler B in the T_n+1 period is zero IOPS. Therefore, a total quantity of resources that are consumed by the user A requests in the T_n+1 period is C_A(T_n+1) = C_A,A(T_n+1) + C_A,B(T_n+1), where C_A(T_n+1)=5 IOPSs. Likewise, the coordinator B acquires that a total quantity of the resources that are consumed by the user B requests in the T_n+1 period is C_B(T_n+1)=C_B,A(T_n+1) + C_B,B(T_n+1), where C_B(T_n+1)=9 IOPSs. The coordinator A sends a message to each of the scheduler A and the scheduler B, where the message carries C_A(T_n+1), so that the scheduler A and the scheduler B acquire, from the coordinator A, that the quantity of IOPSs that are consumed by the user A requests in the T_n+1 period is 5. Likewise, the coordinator B sends a message to each of the scheduler A and the scheduler B, where the message carries C_B(T_n+1), so that the scheduler A and the scheduler B acquire, from the coordinator B, that the quantity of IOPSs that are consumed by the user B requests in the T_n+1 period is 9.

[0050] As shown in FIG. 8, in a T_n+2 period, the scheduler A receives the first user A request P⁶_A,A, where a virtual start time of the P⁶_A,A is S(P⁶_A,A)=max{v(P⁶_A,A), F(P⁵_A,A)}, where v(P⁶_A,A)=F(P⁵_A,A)=2.5, and therefore, S(P⁶_A,A)=2.5.

where

and C_A,B(T_n+1)=0 and C_A,A(T_n+1)=5 in the T_n+1 period; then d(P_A,A(T_n+1))=0; c(P⁶_A,A)=1, φ_A=2, and therefore, F(P⁶_A,A)=3.

[0051] Likewise, a virtual start time of a P⁷_A,A is 3, and a virtual finish time of the P⁷_A,A is 3.5; a virtual start time of a P⁸_A,A is 3.5, and a virtual finish time of the P⁸_A,A is 4; and a virtual start time of a P⁸_A,A is 4, and a virtual finish time of the P⁹_A,A is 4.5. In the T_n+2 period, the scheduler A receives the first user B request P⁵_B,A, where S(P⁵_B,A)=max{v(P⁵_B,A), F(P⁴_B,A)}, v(P⁵_B,A)=F(P⁴_B,A)=4, and therefore, S(P⁵_B,A)=4. A virtual finish time of the P⁵_B,A is

where

C_B,B(T_n+1)=5, N_B,A(T_n+1)=4, d(P_B,A(T_n+2))=1.25, φ_B=1, and therefore, F(P⁵_B,A)=6.25. Likewise, a virtual start time of a P⁶_B,A is 6.25, and a virtual finish time of the P⁶_B,A is 8.5; a virtual start time of a P⁷_B,A is 8.5, and a virtual finish time of the P⁷_B,A is 10.75; and a virtual start time of a P⁸_B,A is 10.75, and a virtual finish time of the P⁸_B,A is 13.

[0052] The scheduler A computes virtual start times and virtual finish times of the P⁶_A,A, P⁷_A,A, P⁸_A,A, and P⁹_A,A, and the scheduler A computes virtual start times and virtual finish times of the P⁵_B,A, P⁶_B,A, P⁷_B,A, and P⁸_B,A. According to the virtual start times of the user requests that are received by the scheduler A, the user requests are ranked in a scheduling queue in ascending order of the virtual start times for processing, that is, an order from the queue head to the queue tail in the scheduling queue is P⁶_A,A, P⁷_A,A, P⁸_A,A, P⁹_A,A, P⁵_B,A, P⁶_B,A, P⁷_B,A, and P⁸_B,A, as shown in FIG. 9. The storage array A processes the user requests in the order of P⁶_A,A, P⁷_A,A, P⁸_A,A, P⁹_A,A, P⁵_B,A, P⁶_B,A, P⁷_B,A, and P⁸_B,A.

[0053] In the T_n+2 period, the scheduler B receives the first user A request P¹_A,B and the first user B request P⁶_B,B. It is assumed that the scheduler B receives the P¹_A,B and the P⁶_B,B at a same moment. A virtual start time of the P⁶_B,B is S(P⁶_B,B)=max{v(P⁶_B,B), F(P⁵_B,B)}, where v(P⁶_B,B)=F(P⁵_B,B)=5, and therefore, S(P⁶_B,B)=5. A virtual finish time of the P⁶_B,B is

where c(P⁶_B,B)=1,

and φ_B=1, and therefore, F(P⁶_B,B)=6.8. Likewise, a virtual start time of a P⁷_B,B is 6.8, and a virtual finish time of the P⁷_B,B is 8.6; a virtual start time of a P⁸_B,B is 8.6, and a virtual finish time of the P⁸_B,B is 10.4. A virtual start time of the P¹_A,B is S(P¹_A,B)=max{v(P¹_A,B), F(P⁰_A,B)}, where F(P⁰_A,B)=0 and v(P¹_A,B)=v(P⁶_B,B)=5, and therefore, S(P¹_A,B)=5. A virtual finish time of the P¹_A,B is

where c(P¹_A,B)=1,

and N_A,A(T_n+1)=0; then d(P_A,B(T_n+2))=0, and φ_A=2, and therefore, F(P¹_A,B)=5.5. Likewise, a virtual start time of a P²_A,B is 5.5, and a virtual finish time of the P²_A,B is 6; and a virtual start time of a P³_A,B is 6, and a virtual finish time of the P³_A,B is 6.5.

[0054] The scheduler B adds the user requests to a scheduling queue, and the user requests are ranked in the scheduling queue in ascending order of the virtual start times, that is, an order from the queue head to the queue tail in the scheduling queue is P¹_A,B, P⁶_B,B, P²_A,B, P³_A,B, P⁷_B,B, and P⁸_B,B, as shown in FIG. 10. The storage array B processes the user requests in the order of P¹_A,B, P⁶_B,B, P²_A,B, P³_A,B, P⁷_B,B, and P⁸_B,B.

[0055] The method for scheduling a user request in a distributed resource system provided by this embodiment of the present invention may be further applied to a scenario shown in FIG. 11, where the distributed resource system includes a user 1 to a user H, a scheduler S₁ to a scheduler S_M, a resource-providing entity R₁ to a resource-providing entity R_M, and a coordinator G₁ to a coordinator G_Y, where H is a natural number greater than or equal to 2, M is a natural number greater than or equal to 2, and N is a natural number. A scheduler S_x communicates with a resource-providing entity R_x, where x is any natural number from 1 to M. Any user z can send a user z request to any scheduler S_x. The R_x provides a resource for a user request received by the S_x. The coordinator G_y communicates with any R_x, where y is any natural number from 1 to Y, and a value of Y may be less than or equal to M, or may be greater than M. Y may also be 1, that is, there is only one coordinator in the distributed resource system. A quantity of coordinators may be determined according to a quantity of schedulers, or may be determined according to a quantity of users, which is not limited in this embodiment of the present invention. Any user z sends the i^th user z request to a scheduler S_d in a T_n+1 period, where i is a natural number, d is a natural number, 1≤d≤M, and S_d indicates one of S₁ to S_M. When the S_d needs to schedule the received user request, the following steps are performed:

Step 1201. The S_d acquires, in the T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by the user z request in a T_n period, where

d and k are natural numbers, 1≤d≤M, and 1≤k≤Y; a resource weight of the user z is φ_z ; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; z indicates an identifier of the user; and

indicates the sum of resources that are provided by the R₁ to the R_M and consumed by user z requests received by the S₁ to the S_M in the T_n period.

[0056] In step 1201, C_z,x(T_n) specifically includes the sum of resources that are provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period. N_z,x(T_n) indicates a quantity of user z requests that are received by the S_x in the T_n period. The S_x sends C_z,x(T_n) to a G_k, and the G_k acquires C_z(T_n) according to

A manner of acquiring C_z(T_n) is that the S_x proactively sends C_z,x(T_n) to the G_k. Another manner is that the S_x receives a G_k request, and sends C_z,x(T) to the G_k according to the G_k request. The G_k acquires C_z,x(T_n) according to

The S_x stores N_z,x(T_n) and C_z,x(T_n).

[0057] Step 1202. The S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm, where the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

[0058] The S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), the Pⁱ_z,d by using a scheduling algorithm, thereby implementing global scheduling on the user z request and ensuring a performance requirement of the user z. Step 1202 specifically includes:

That the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), the Pⁱ_z,d by using a scheduling algorithm specifically includes that: the S_d computes a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), and adds the Pⁱ_z,d to a scheduling queue, where the scheduling queue ranks the user request according to a value of the virtual start time of the user request. S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(P^i-1_z,d)},

and

where v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d; or, S(Pⁱ_z,d)max{v(Pⁱ_z,d),



and

where v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

[0059] For meanings of the virtual start time S(Pⁱ_A,A), the virtual finish time F(Pⁱ_A,A), and the virtual time v(Pⁱ_A,A), refer to a start-time fairness queuing (Start-time fairness queuing, SFQ) algorithm, and details are not described in this embodiment of the present invention again. d(P_z,d(T_n+1)) indicates a value of a delay of each Pⁱ_z,d received by the S_d in the T_n+1 period.

[0060] When either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0. c(Pⁱ_z,d) may be specifically an IOPS or bandwidth that is consumed by the Pⁱ_z,d. Step 1201 specifically includes:

the G_k is determined according to z%Y=k, where z%Y indicates that a modulo operation is performed on z and Y; or the G_k is determined according to Hash(z)%Y=k, where Hash(z) indicates that z is computed by using a hash function, and Hash(z)%Y indicates that a modulo operation is performed on a value and Y, where the value is obtained by computing z by using the hash function.

[0061] In the application scenario shown in FIG. 11, the C_K is configured to send C_z(T_n) to each scheduler S_x.

[0062] In the application scenario shown in FIG. 11, the method that the S_d uses a scheduling algorithm to schedule the user z request is applicable to scheduling of any user request in the distributed resource system in this embodiment of the present invention. Requests of two or more users may be scheduled in the S_d.

[0063] In the application scenario shown in FIG. 11, the scheduler S_d schedules, by using a scheduling algorithm, the user z request Pⁱ_z,d according to C_z(T_n), C_z,d(T_n), and N_z,d(T_n) of the user z in the distributed resource system in the T_n period, and with reference to the weight φ_z of the user, thereby implementing global scheduling on the user z request and ensuring a performance requirement of the user z.

[0064] In a system architecture in this embodiment of the present invention, the user z request can be scheduled without depending on a user agent.

[0065] In this embodiment of the present invention, that a resource is an IOPS is used as an example, and generally, one user request consumes one IOPS; or that a resource is network bandwidth is used as an example, and a resource that is consumed by one user request is network bandwidth for the user request. When a resource consumed by a user request is network bandwidth, a size of a resource that is provided by a resource-providing entity and consumed by each user request is determined by the user request.

[0066] In this embodiment of the present invention, scheduling a user request by using a scheduling algorithm includes: computing a virtual start time and a virtual finish time of the user request according to the scheduling algorithm in this embodiment of the present invention; and adding the user request to a scheduling queue, where the user request is ranked in the scheduling queue according to a value of the virtual start time of the user request. When it is not required to schedule the user request, the virtual start time and the virtual finish time of the user request need to be computed according to the scheduling algorithm in this embodiment of the present invention, and the user request does not need to be added to a scheduling queue for ranking.

[0067] In this embodiment of the present invention, S_d, S_x, R_x, G_y, and G_k are merely used to mark a specific device each, where the device herein may be a physical device or a logical device. Likewise, a device may also be expressed in a manner of first, second, and the like. There is only a difference in specific expression manners, which does not impose a limitation on the solution of the present invention.

[0068] As shown in FIG. 12, an embodiment of the present invention provides a scheduler S_d. The scheduler S_d is applied to a distributed resource system, where the distributed resource system includes schedulers S_x, resource-providing entities R_x and a coordinator G_y, where x is consecutive natural numbers from 1 to M, and M≥2; d is a number in x; y is consecutive natural numbers from 1 to Y; the S_x communicates with the R_x; a C_y communicates with any S_x; and the scheduler S_d includes an acquiring unit 1301 and a scheduling unit 1302, where:

the acquiring unit 1301 is configured to acquire, in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, where

k is a natural number, and 1≤k≤Y; a resource weight of the user z is φ_z ; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

the scheduling unit 1302 is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm, where the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

[0069] Specifically, that the scheduling unit 1302 is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm specifically includes:

computing a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n) ; and adding the Pⁱ_z,d to a scheduling queue, where the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

[0070] Specifically, S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(Pⁱ_z,d)},

and

where v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

[0071] Specifically, S(Pⁱ_z,d)=max{v(Pⁱ_z,d),



and

where v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

[0072] For meanings of the virtual start time S(Pⁱ_A,A), the virtual finish time F(Pⁱ_A,A), and the virtual time v(Pⁱ_A,A), refer to a start-time fairness queuing (Start-time fairness queuing, SFQ) algorithm, and details are not described in this embodiment of the present invention again. Specifically, when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

[0073] The S_d schedules, according to φ_z, C_z(T_n)), C_z,d(T_n), the Pⁱ_z,d by using the scheduling algorithm. The user z request can be scheduled without depending on a user agent. In addition, the S_d schedules, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), the Pⁱ_z,d by using the scheduling algorithm, thereby implementing global scheduling on the user z request and ensuring a performance requirement of the user z.

[0074] In an implementation manner of another resource scheduling method in a distributed resource system, the distributed resource system includes multiple schedulers, where a first scheduler in the multiple schedulers acquires, from a coordinator of the first user, the sum of resources that are consumed by a user request of the first user in the multiple schedulers in a previous period; and

the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed by the user request of the first user in the multiple schedulers in the previous period, a resource that is consumed by the user request of the first user in the first scheduler in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

[0075] Further, that the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed by the user request of the first user in the multiple schedulers in the previous period, a resource that is consumed by the user request of the first user in the first scheduler in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period specifically includes that:

the first scheduler computes a virtual start time and a virtual finish time of the user request of the first user according to the resource weight of the first user, the sum of resources that are consumed by the user request of the first user in the multiple schedulers in the previous period, the resource that is consumed by the user request of the first user in the first scheduler in the previous period, and the quantity of user requests of the first user received by the first scheduler in the previous period; and adds the user request of the first user to a scheduling queue, where the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

[0076] In another distributed resource system, the distributed resource system includes multiple schedulers. As shown in FIG. 12, a first scheduler in the multiple schedulers includes an acquiring unit 1301 and a scheduling unit 1302, where:

the first acquiring unit 1301 is configured to acquire, from a coordinator of the first user, the sum of resources that are consumed by a user request of the first user in the multiple schedulers in a previous period; and

the scheduling unit 1302 is configured to schedule the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed by the user request of the first user in the multiple schedulers in the previous period, a resource that is consumed by the user request of the first user in the first scheduler in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

[0077] Further, the scheduling unit 1302 is specifically configured to compute a virtual start time and a virtual finish time of the user request of the first user according to the resource weight of the first user, the sum of resources that are consumed by the user request of the first user in the multiple schedulers in the previous period, the resource that is consumed by the user request of the first user in the first scheduler in the previous period, and the quantity of user requests of the first user received by the first scheduler in the previous period; and add the user request of the first user to a scheduling queue, where the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

[0078] The technical solutions provided by the embodiments of the present invention may also be applied to another scenario. For example, a distributed resource system is a distributed computing system, and a resource-providing entity provides a computing resource for a user request; or, a distributed resource system may be a distributed network system, and a resource-providing entity provides network bandwidth for a user request; or, a resource-providing entity may further provide a memory resource for a user request.

[0079] A person of ordinary skill in the art may be aware that, in combination with the examples described in the embodiments disclosed in this specification, units and algorithm steps may be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed by hardware or software depends on particular applications and design constraint conditions of the technical solutions. A person skilled in the art may use different methods to implement the described functions for each particular application, but it should not be considered that the implementation goes beyond the scope of the present invention.

[0080] It may be clearly understood by a person skilled in the art that, for the purpose of convenient and brief description, for a detailed working process of the foregoing system, apparatus, and unit, refer to a corresponding process in the foregoing method embodiments, and details are not described herein again.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed system and method may be implemented in other manners. For example, the described apparatus embodiment is merely exemplary. For example, the unit division is merely logical function division and may be other division in actual implementation. For example, a plurality of units or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections may be implemented through some interfaces. The indirect couplings or communication connections between the apparatuses or units may be implemented in electronic, mechanical, or other forms.

[0082] The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

[0083] In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each of the units may exist alone physically, or two or more units are integrated into one unit.

[0084] When the functions are implemented in the form of a software functional unit and sold or used as an independent product, the functions may be stored in a computer-readable non-volatile storage medium. Based on such an understanding, the technical solutions of the present invention essentially, or the part contributing to the prior art, or some of the technical solutions may be implemented in a form of a software product. The software product is stored in a non-volatile storage medium, and includes several instructions for instructing a computer device (which may be a personal computer, a server, or a network device) to perform all or some of the steps of the methods described in the embodiments of the present invention. The foregoing non-volatile storage medium includes: any medium that can store program code, such as a USB flash drive, a removable hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disc.

Claims

1. A method for scheduling a user request in a distributed resource system, wherein the distributed resource system comprises schedulers S_x, resource-providing entities R_x, and a coordinator G_y, wherein x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x; the G_y communicates with any S_x; and the method comprises:

acquiring, by an S_d in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, wherein

d and k are natural numbers, 1≤d≤M, and 1≤k≤Y; a resource weight of the user z is φ_z ; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

scheduling, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T), a Pⁱ_z,d by using a scheduling algorithm, wherein the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

2. The method according to claim 1, wherein the scheduling, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm specifically comprises:

computing, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d ; and adding the Pⁱ_z,d to a scheduling queue, wherein the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

3. The method according to claim 2, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(P^i-1_z,d)},

and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

4. The method according to claim 3, wherein when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

5. The method according to claim 2, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d),



and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

6. The method according to claim 5, wherein when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+))=0.

7. The method according to claim 1, wherein the G_k is determined according to z%Y=k, wherein z%Y indicates z modulo Y.

8. The method according to claim 1, wherein the G_k is determined according to Hash(z)%Y=k, wherein Hash(z) indicates that z is computed by using a hash function, and Hash(z)%Y indicates a value obtained by computing z by using the hash function modulo Y.

9. A scheduler S_d, wherein the scheduler S_d is applied to a distributed resource system, and the distributed resource system comprises schedulers S_x, resource-providing entities R_x, and a coordinator G_y, wherein x is consecutive natural numbers from 1 to M, and M≥2; d is a number in x; y is consecutive natural numbers from 1 to Y; the S_x communicates with the R_x ; a C_y communicates with any S_x ; the scheduler S_d comprises a central processing unit and a memory, wherein the central processing unit executes an executable instruction in the memory, to perform the following steps:

acquiring, in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T_n) that is consumed by a user z request in a T_n period, wherein

k is a natural number, and 1≤k≤Y; a resource weight of the user z is φ_z; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

scheduling, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm, wherein the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

10. The scheduler S_d according to claim 9, wherein the scheduling, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm specifically comprises:

computing, by the S_d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d ; and adding the Pⁱ_z,d to a scheduling queue, wherein the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

11. The scheduler S_d according to claim 10, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(P^i-1_z,d)}

and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

12. The scheduler S_d according to claim 11, wherein when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

13. The scheduler S_d according to claim 10, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d),

and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

14. The scheduler S_d according to claim 11, wherein when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

15. A scheduler S_d, wherein the scheduler S_d is applied to a distributed resource system, and the distributed resource system comprises schedulers S_x, resource-providing entities R_x, and a coordinator G_y, wherein x is consecutive natural numbers from 1 to M, and M≥2; d is a number in x; y is consecutive natural numbers from 1 to Y; the S_x communicates with the R_x ; a C_y communicates with any S_x ; and the scheduler S_d comprises an acquiring unit and a scheduling unit, wherein:

the acquiring unit is configured to acquire, in a T_n+1 period and from a coordinator G_k of a user z, a resource C_z(T) that is consumed by a user z request in a T_n period, wherein

k is a natural number, and 1≤k≤Y; a resource weight of the user z is φ_z ; C_z,x(T) is a resource that is provided by the R_x and consumed by N_z,x(T) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and

the scheduling unit is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T), a Pⁱ_z,d by using a scheduling algorithm, wherein the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

16. The scheduler S_d according to claim 15, wherein that the scheduling unit is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T), and N_z,d(T), a Pⁱ_z,d by using a scheduling algorithm specifically comprises:

computing a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_,d) of the Pⁱ_z,d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n); and adding the Pⁱ_z,d to a scheduling queue, wherein the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

17. The scheduler S_d according to claim 16, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(P^i-1_z,d)}

and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

18. The scheduler S_d according to claim 17, wherein when either of C_z(T_n) - C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

19. The scheduler S_d according to claim 16, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d),



and

wherein v(Pⁱ_,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

20. The scheduler S_d according to claim 18, wherein when either of C_z(T_n)-C_z,d(T_n) and N_z,d(T_n) is 0, d(P_z,d(T_n+1))=0.

21. A non-volatile computer readable storage medium, wherein the non-volatile computer readable storage medium stores a computer instruction that is used to perform user request scheduling in a distributed resource system, and the distributed resource system comprises schedulers S_x, resource-providing entities R_x, and a coordinator G_y, wherein x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x, and a C_y communicates with any S_x ; and a scheduler S_d executes the computer instruction, to perform the method according to any one of claims 1 to 8, wherein d is a number in x.

22. A distributed resource system, wherein the distributed resource system comprises schedulers S_x, resource-providing entities R_x, and a coordinator G_y, wherein x is consecutive natural numbers from 1 to M, and M≥2; y is consecutive natural numbers from 1 to Y, and Y is a natural number; the S_x communicates with the R_x ; and the G_y communicates with any S_x ;
a coordinator G_k of a user z is configured to provide, for a scheduler S_d in a T_n+1 period, a resource C_z(T_n) that is consumed by a user z request in a T_n period, wherein

d and k are natural numbers, 1≤d≤M, and 1≤k≤Y; a resource weight of the user z is φ_z ; C_z,x(T_n) is a resource that is provided by the R_x and consumed by N_z,x(T_n) user z requests received by the S_x in the T_n period; and z indicates an identifier of the user; and
the scheduler S_d is configured to schedule, according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), a Pⁱ_z,d by using a scheduling algorithm, wherein the Pⁱ_z,d is the i^th user z request received by the S_d, and C_z,d(T_n) is a resource that is provided by an R_d and consumed by N_z,d(T_n) user z requests received by the S_d in the T_n period.

23. The distributed resource system according to claim 22, wherein the scheduler S_d is specifically configured to compute a virtual start time S(Pⁱ_z,d) and a virtual finish time F(Pⁱ_z,d) of the Pⁱ_z,d according to φ_z, C_z(T_n), C_z,d(T_n), and N_z,d(T_n), and add the Pⁱ_z,d to a scheduling queue, wherein the scheduling queue ranks the user request according to a value of the virtual start time of the user request.

24. The method according to claim 23, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d), F(P^i-1_z,d)},

and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

25. The method according to claim 23, wherein S(Pⁱ_z,d)=max{v(Pⁱ_z,d),



and

wherein v(Pⁱ_z,d) indicates a virtual time of the S_d when the S_d receives the Pⁱ_z,d, and c(Pⁱ_z,d) indicates a resource that is provided by the R_d and consumed by the Pⁱ_z,d.

26. A resource scheduling method in a distributed resource system, wherein the distributed resource system comprises multiple schedulers;

a first scheduler in the multiple schedulers acquires, from a coordinator of the first user, the sum of resources that are consumed in the multiple schedulers by a user request of the first user in a previous period; and

the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

27. The method according to claim 26, wherein that the first scheduler schedules the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period specifically comprises:

computing, by the first scheduler according to the resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, the resource that is consumed in the first scheduler by the user request of the first user in the previous period, and the quantity of the user requests of the first user received by the first scheduler in the previous period, a virtual start time and a virtual finish time of the user request of the first user; and adding the user request of the first user to a scheduling queue, wherein the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

28. A distributed resource system, wherein the distributed resource system comprises multiple schedulers, and a first scheduler in the multiple schedulers comprises an acquiring unit and a scheduling unit, wherein:

the acquiring unit is configured to acquire, from a coordinator of the first user, the sum of resources that are consumed in the multiple schedulers by a user request of the first user in a previous period; and

the scheduling unit is configured to schedule the user request of the first user according to a resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, a resource that is consumed in the first scheduler by the user request of the first user in the previous period, and a quantity of user requests of the first user received by the first scheduler in the previous period.

29. The distributed resource system according to claim 28, wherein the scheduling unit is specifically configured to compute, according to the resource weight of the first user, the sum of resources that are consumed in the multiple schedulers by the user request of the first user in the previous period, the resource that is consumed in the first scheduler by the user request of the first user in the previous period, and the quantity of the user requests of the first user received by the first scheduler in the previous period, a virtual start time and a virtual finish time of the user request of the first user; and add the user request of the first user to a scheduling queue, wherein the scheduling queue ranks the user request of the first user according to a value of the virtual start time of the user request.

Drawing